Technology Licensing Update: New Rules on AI-Related Technology Licensing Published — Key Takeaways
China’s Ministry of Commerce (MOFCOM) and the Ministry of Science and Technology have published the updated Catalog of Technologies Prohibited or Restricted from Export (中国禁止出口限制出口技术目录, zhōngguó jìnzhǐ chūkǒu xiànzhì chūkǒu jìshù mùlù), adding 7 new AI-related categories and revising 5 existing ones, bringing the total regulated items to 164. For the first time, technologies in generative AI, AI-assisted content generation, and high-performance AI chips are explicitly listed under restricted export licensing, meaning any cross-border license or transfer now requires MOFCOM approval. This update applies immediately to all technology licensing agreements involving Chinese entities, including joint-ventures and wholly foreign-owned enterprises (外商独资企业, WFOE, wàishāng dúzī qǐyè).
Background: Why New Rules on AI Tech Licensing
The update, published on April 15, 2025, replaces the previous catalog from December 2023 and responds to rapid advances in generative AI, autonomous systems, and semiconductor design. MOFCOM’s goal is to prevent the transfer of dual-use AI capabilities that could serve military or national security applications, following similar moves by the United States (export controls on AI chips) and the European Union (AI Act).
In 2023, China exported approximately ¥680 billion in technology-related services under license agreements, a 14% increase from 2022. Of that, AI-related licensing accounted for an estimated ¥92 billion, but under the new rules, at least 30–40% of these agreements will now require mandatory government review. The timeline is tight: existing agreements must be re-filed within 90 days of the catalog’s publication, which means foreign companies have until mid-July 2025 to comply or face penalties.
Key Changes in the Updated Catalog
The most significant shift is the inclusion of generative AI model architecture, training datasets exceeding 100 TB, and AI inference optimization techniques under the “restricted” category. These are now classified under new item codes R-05-01 through R-05-07. Previously, only narrow AI applications like facial recognition were listed. Now, any technology license that involves “AI model fine-tuning, distillation, or pruning” as part of a cross-border transfer will require a technology licensing contract approval from the local MOFCOM branch.
Additionally, the updated catalog introduces a new “prohibited” category for AI technologies used in autonomous weapons systems and mass surveillance, matching similar restrictions in Japan and the UK. In total, 24 items are now classified as prohibited (up from 18 in 2023) and 140 items as restricted (up from 113). For companies operating in AI development, manufacturing, or data processing, this means their standard technology licensing agreements (技术许可合同, jìshù xǔkě hétong) for products like AI chips or cloud-based model APIs may no longer be eligible for simple registration and instead require a full export licensing review that takes 45–90 business days.
Table: Comparison of AI-Related Items — 2023 vs 2025 Catalog
| Item Category | 2023 Catalog Count | 2025 Catalog Count | Key Additions | Licensing Impact |
|---|---|---|---|---|
| Generative AI Model Technologies | 0 | 7 | LLM architecture, training data >100TB, RLHF methods | Restricted — MOFCOM approval required |
| AI Chip Design & Manufacturing | 3 | 6 | AI accelerator design, chiplet packaging for AI | Restricted — approved licensing only |
| AI-Based Content Generation | 0 | 4 | Text, image, video, and audio generation models | Restricted — government review per agreement |
| Autonomous Systems & Weapons | 2 | 4 | AI-driven drone swarms, targeting algorithms | Prohibited — no cross-border transfer allowed |
| AI Data Processing & Augmentation | 1 | 3 | Synthetic data generation, edge AI training | Restricted — licensing contract must be filed |
Impact on Foreign Enterprises and Licensing Strategies
For foreign companies licensing AI technology into or out of China, the new rules create two distinct compliance pathways. If the technology appears in the “restricted” category, a Technology Import and Export Contract License (技术进出口合同许可证, jìshù jìnchūkǒu hétong xǔkězhèng) must be obtained. This requires submitting a detailed technical description, end-user certificates, and a security assessment. If the technology is listed as “prohibited,” no licensing is permitted under any circumstance — even for internal use within a WFOE (外商独资企业, WFOE, wàishāng dúzī qǐyè), unless an exemption is granted by the State Council.
For example, a U.S.-based AI startup licensing its generative AI model to a Chinese joint venture must now assess whether the model’s architecture is covered by item R-05-01. If yes, the license agreement must be filed with MOFCOM, and government approval can take up to 180 days if a national security review is triggered. In contrast, a European company licensing AI-based image recognition software for factory automation may fall outside the catalog if the model size stays below 100 TB and no fine-tuning data is transferred — but the burden of proof now lies with the licensor.
Compliance Requirements and Timelines
All existing AI-related technology licensing agreements signed before April 15, 2025 must be reviewed and, if necessary, amended within 90 days to reflect the new catalog. MOFCOM has given a one-time grace period: companies that voluntarily re-file and correct their agreements before the deadline face no fines. After the deadline, non-compliance penalties range from ¥100,000 to ¥5 million per violation, plus potential suspension of technology import/export privileges.
The process involves three steps: (i) self-classify your technology against the updated catalog, (ii) prepare a technology export application with a Chinese-language description and end-user undertaking, and (iii) submit to the local MOFCOM office in the province where the Chinese entity is registered. In practice, 80% of applications in the AI category take 60–90 business days to process, so early submission is critical. For companies with multiple licensing agreements, each agreement requires a separate application.
Decision Framework for AI Licensing in China
Given the new rules, foreign companies must choose the right compliance path based on their technology and agreement structure:
- If your technology is on the restricted list (R-05-01 to R-05-07): Obtain a Technology Import and Export Contract License from MOFCOM before signing or renewing any licensing agreement. Budget 45–90 business days for approval. Example: licensing a generative AI model to a Chinese partner.
- If your technology is on the prohibited list (P-02-01 to P-02-04): Do not proceed with cross-border licensing. Instead, consider restructuring the agreement as a remote service (API-based access), which may fall outside the licensing scope — but only after legal review. Example: AI targeting algorithms for defense applications.
- If your technology is not explicitly listed: File a voluntary classification request with MOFCOM to obtain a written confirmation. This protects you from future audits. Example: AI-based inventory optimization software with no generative AI components.
3 Pitfalls in AI Technology Licensing Under the New Rules
NEXT STEPS
- Conduct an AI Technology Classification Audit — Inventory all your existing and planned licensing agreements involving AI, chips, or data processing. Map each item to the new catalog codes. Read our guide: AI Technology Export Classification in China (2025)
- Re-file Existing Licensing Agreements — If any agreement includes a restricted AI technology, submit an amendment and new license application before the 90-day deadline. Use our technology licensing compliance service
- Review Joint Venture and WFOE Structures — For companies using a WFOE (外商独资企业) to manage AI R&D in China, ensure internal technology transfers do not inadvertently trigger export licensing requirements. Read our comparison: WFOE vs. JV for AI Licensing in China
— China Gateway 360 —
Remote China market entry support, built around execution.
